Welcome back!

Ilya Sutskever, former Chief Researcher of OpenAI, has launched an ambitious new venture: Safe Superintelligence Inc (SSI). The aim? To rapidly develop superintelligent AI systems with a strong focus on safety and insulation from commercial pressures. Let’s unpack this…

Today’s Summary:

• Ilya Sutskever launches Safe Superintelligence Inc

• Claude 3.5 Sonnet released, beats GPT-4o on benchmarks

• Microsoft releases Florence-2 vision model

• McDonald's removes AI drive-throughs

• OpenAI halts AI political candidates

• Dell partners with Nvidia for xAI factory

• 2 new tools

TOP STORY

OpenAI Co-Founder Ilya Sutskever Launches Safe Superintelligence Inc

The Summary: Ilya Sutskever, former OpenAI chief scientist and co-founder, has launched a new AI Research Lab called Safe Superintelligence Inc (SSI). The startup aims to develop superintelligent AI systems with a focus on safety. Sutskever considers this the most critical technical challenge of our time.

Unlike OpenAI, SSI is committed to advancing AI capabilities quickly while ensuring safety remains the top priority, free from commercial pressures. Joining him in this venture are Daniel Gross, an investor formerly with Apple and Daniel Levy, a researcher formerly with OpenAI.

Key details:

• Headquarters in Palo Alto and Tel Aviv

• Mission: develop safe superintelligent AI

• Safety and capabilities as parallel challenges

• Rapid scaling with a safety focus

• Free from management and product cycle distractions

• Shielded from short-term commercial interests

Why it matters: Sutskever launch of SSI marks a new step in AI development, emphasizing safety. As AI grows more powerful, addressing the risks of superintelligence is crucial. SSI's unique approach, focusing solely on safety and insulated from profit motives, could lead to new groundbreaking work.

MODELS

Anthropic Releases Claude 3.5 Sonnet, Beats GPT-4o on Benchmarks

The Summary: Anthropic has launched Claude 3.5 Sonnet, a new AI model that outperforms its predecessors and GPT-4o on various benchmarks. The model offers improved text and image analysis capabilities, along with faster processing speeds.

Anthropic also introduced Artifacts, a new workspace for editing AI-generated content. These improvements represent a new incremental step in AI technology.

Key details:

• Beats GPT-4o on several benchmarks

• Outperforms previous Anthropic models

• Analyzes text and images and generates text

• Twice the speed of the previous Claude 3 Opus model

• Context window of 200,000 tokens (vs 128K for GPT-4o)

• Artifacts is a new workspace for editing AI-generated content

• Available now for free through claude.ai, iOS app, and API

• An even better version, Claude 3.5 Opus, will be released soon with additional features such as web search

Why it matters: Claude 3.5 Sonnet offers notable improvements in speed and capability. This release highlights the current state of AI progress, showing that improvements are now more incremental. It also demonstrates Anthropic strategy to compete in the AI market by offering better performance at competitive prices. The introduction of Artifacts suggests a focus on building an agentic ecosystem around AI models.

MODELS

Microsoft Releases Florence-2 Vision Model

The Summary: Microsoft has released Florence-2, an open-source AI model that can handle a variety of vision tasks using a unified representation. Trained on 5.4 billion annotations across 126 M images, Florence-2 understands images at multiple levels, from high-level concepts to detailed attributes.

In contrast to traditional computer vision models specialized for single tasks, Florence-2 demonstrates remarkable versatility and outperforms much larger models on captioning, object detection, grounding, and segmentation without additional training data for those tasks. It provides great accuracy and speed. You can try it here.

Key details:

• Handles captioning, object detection, region captioning, region proposal, region segmentation, OCR, and more

• Uses a unified architecture that combines vision and language

• Achieves new best performance at labeling visual regions

• As a generalist model, it sets competitive records after fine-tuning

• Provides an efficient multipurpose vision system, boosting many computer vision tasks

Why it matters: Florence-2 deeply understands images in a unified way, grasping high-level meanings as well as intricate details. This versatile system can be used in many vision applications out-of-the-box and adapted easily. Its efficiency shows the promise of unified models over traditional task-specific ones.

QUICK NEWS

Quick news

• Dell and Nvidia to build an AI factory for Elon Musk’s xAI

• McDonalds removes AI drive-throughs

• OpenAI shuts down access to Mayor AI candidates

TOOLS

🥇 New tools

• Rtranslator - Free realtime translator app for Android

• AI Logo Reveals - Animate any logo